1
0
mirror of https://github.com/cc65/cc65.git synced 2024-11-18 00:07:21 +00:00
cc65/libsrc/common/_scanf.c

869 lines
27 KiB
C

/*
** _scanf.c
**
** (c) Copyright 2001-2005, Ullrich von Bassewitz <uz@cc65.org>
** 2014-09-10, Greg King <greg.king5@verizon.net>
**
** This is the basic layer for all scanf-type functions. It should be
** rewritten in assembly, at some time in the future. So, some of the code
** is not as elegant as it could be.
*/
#include <stddef.h>
#include <stdarg.h>
#include <stdbool.h>
#include <stdio.h>
#include <string.h>
#include <setjmp.h>
#include <limits.h>
#include <errno.h>
#include <ctype.h>
/* _scanf() can give EOF to these functions. But, the macroes can't
** understand it; so, they are removed.
*/
#undef isspace
#undef isxdigit
#include "_scanf.h"
#pragma static-locals(on)
/*****************************************************************************/
/* SetJmp return codes */
/*****************************************************************************/
enum {
RC_OK, /* setjmp() call */
RC_NOCONV, /* No conversion possible */
RC_EOF /* EOF reached */
};
/*****************************************************************************/
/* Data */
/*****************************************************************************/
static const char* format; /* Copy of function argument */
static const struct scanfdata* D_; /* Copy of function argument */
static va_list ap; /* Copy of function argument */
static jmp_buf JumpBuf; /* "Label" that is used for failures */
static char F; /* Character from format string */
static unsigned CharCount; /* Characters read so far */
static int C; /* Character from input */
static unsigned Width; /* Maximum field width */
static long IntVal; /* Converted int value */
static int Assignments; /* Number of assignments */
static unsigned char IntBytes; /* Number of bytes-1 for int conversions */
/* Flags */
static bool Converted; /* Some object was converted */
static bool Positive; /* Flag for positive value */
static bool NoAssign; /* Suppress assignment */
static bool Invert; /* Do we need to invert the charset? */
static unsigned char CharSet[(1+UCHAR_MAX)/CHAR_BIT];
static const unsigned char Bits[CHAR_BIT] = {
0x01, 0x02, 0x04, 0x08, 0x10, 0x20, 0x40, 0x80
};
/* We need C to be 16 bits since we cannot check for EOF otherwise.
** Unfortunately, this causes the code to be quite larger, even if for most
** purposes, checking the low byte would be enough, since if C is EOF, the
** low byte will not match any useful character anyway (at least for the
** supported platforms - I know that this is not portable). So the following
** macro is used to access just the low byte of C.
*/
#define CHAR(c) (*((unsigned char*)&(c)))
/*****************************************************************************/
/* Character sets */
/*****************************************************************************/
/* We don't want the optimizer to ruin our "perfect" ;-)
** assembly code!
*/
#pragma optimize (push, off)
static unsigned FindBit (void)
/* Locate the character's bit in the charset array.
** < .A - Argument character
** > .X - Offset of the byte in the character-set mask
** > .A - Bit-mask
*/
{
asm ("pha");
asm ("lsr a"); /* Divide by CHAR_BIT */
asm ("lsr a");
asm ("lsr a");
asm ("tax"); /* Byte's offset */
asm ("pla");
asm ("and #%b", CHAR_BIT-1);
asm ("tay"); /* Bit's offset */
asm ("lda %v,y", Bits);
return (unsigned) __AX__;
}
#pragma optimize (pop)
static void __fastcall__ AddCharToSet (unsigned char /* C */)
/* Set the given bit in the character set */
{
FindBit();
asm ("ora %v,x", CharSet);
asm ("sta %v,x", CharSet);
}
#pragma optimize (push, off)
static unsigned char IsCharInSet (void)
/* Check if the char. is part of the character set. */
{
/* Get the character from C. */
asm ("lda #$00");
asm ("ldx %v+1", C);
asm ("bne L1"); /* EOF never is in the set */
asm ("lda %v", C);
FindBit();
asm ("and %v,x", CharSet);
asm ("L1:");
asm ("ldx #$00");
return (unsigned char) __AX__;
}
#pragma optimize (pop)
static void InvertCharSet (void)
/* Invert the character set */
{
asm ("ldy #%b", sizeof (CharSet) - 1);
asm ("L1:");
asm ("lda %v,y", CharSet);
asm ("eor #$FF");
asm ("sta %v,y", CharSet);
asm ("dey");
asm ("bpl L1");
}
/*****************************************************************************/
/* Code */
/*****************************************************************************/
static void PushBack (void)
/* Push back the last (unused) character, provided it is not EOF. */
{
/* Get the character from C. */
/* Only the high-byte needs to be checked for EOF. */
asm ("ldx %v+1", C);
asm ("bne %g", Done);
asm ("lda %v", C);
/* Put unget()'s first argument on the stack. */
asm ("jsr pushax");
/* Copy D into the zero-page. */
(const struct scanfdata*) __AX__ = D_;
asm ("sta ptr1");
asm ("stx ptr1+1");
/* Copy the unget vector to jmpvec. */
asm ("ldy #%b", offsetof (struct scanfdata, unget));
asm ("lda (ptr1),y");
asm ("sta jmpvec+1");
asm ("iny");
asm ("lda (ptr1),y");
asm ("sta jmpvec+2");
/* Load D->data into __AX__. */
asm ("ldy #%b", offsetof (struct scanfdata, data) + 1);
asm ("lda (ptr1),y");
asm ("tax");
asm ("dey");
asm ("lda (ptr1),y");
/* Call the unget routine. */
asm ("jsr jmpvec");
/* Take back that character's count. */
asm ("lda %v", CharCount);
asm ("bne %g", Yank);
asm ("dec %v+1", CharCount);
Yank:
asm ("dec %v", CharCount);
Done:
;
}
static void ReadChar (void)
/* Get an input character, count characters */
{
/* Move D to ptr1 */
asm ("lda %v", D_);
asm ("ldx %v+1", D_);
asm ("sta ptr1");
asm ("stx ptr1+1");
/* Copy the get vector to jmpvec */
asm ("ldy #%b", offsetof (struct scanfdata, get));
asm ("lda (ptr1),y");
asm ("sta jmpvec+1");
asm ("iny");
asm ("lda (ptr1),y");
asm ("sta jmpvec+2");
/* Load D->data into __AX__ */
asm ("ldy #%b", offsetof (struct scanfdata, data) + 1);
asm ("lda (ptr1),y");
asm ("tax");
asm ("dey");
asm ("lda (ptr1),y");
/* Call the get routine */
asm ("jsr jmpvec");
/* Assign the result to C */
asm ("sta %v", C);
asm ("stx %v+1", C);
/* If C is EOF, don't bump the character counter.
** Only the high-byte needs to be checked.
*/
asm ("inx");
asm ("beq %g", Done);
/* Must bump CharCount. */
asm ("inc %v", CharCount);
asm ("bne %g", Done);
asm ("inc %v+1", CharCount);
Done:
;
}
#pragma optimize (push, off)
static void __fastcall__ Error (unsigned char /* Code */)
/* Does a longjmp using the given code */
{
asm ("pha");
(char*) __AX__ = JumpBuf;
asm ("jsr pushax");
asm ("pla");
asm ("ldx #>$0000");
asm ("jmp %v", longjmp);
}
#pragma optimize (pop)
static void CheckEnd (void)
/* Stop a scan if it prematurely reaches the end of a string or a file. */
{
/* Only the high-byte needs to be checked for EOF. */
asm ("ldx %v+1", C);
asm ("beq %g", Done);
Error (RC_EOF);
Done:
;
}
static void SkipWhite (void)
/* Skip white space in the input and return the first non white character */
{
while ((bool) isspace (C)) {
ReadChar ();
}
}
#pragma optimize (push, off)
static void ReadSign (void)
/* Read an optional sign and skip it. Store 1 in Positive if the value is
** positive, store 0 otherwise.
*/
{
/* We can ignore the high byte of C here, since if it is EOF, the lower
** byte won't match anyway.
*/
asm ("lda %v", C);
asm ("cmp #'-'");
asm ("bne %g", NotNeg);
/* Negative value */
asm ("sta %v", Converted);
asm ("jsr %v", ReadChar);
asm ("lda #$00"); /* Flag as negative */
asm ("beq %g", Store);
/* Positive value */
NotNeg:
asm ("cmp #'+'");
asm ("bne %g", Pos);
asm ("sta %v", Converted);
asm ("jsr %v", ReadChar); /* Skip the + sign */
Pos:
asm ("lda #$01"); /* Flag as positive */
Store:
asm ("sta %v", Positive);
}
#pragma optimize (pop)
static unsigned char __fastcall__ HexVal (char C)
/* Convert a digit to a value */
{
return (bool) isdigit (C) ?
C - '0' :
(char) tolower ((int) C) - ('a' - 10);
}
static void __fastcall__ ReadInt (unsigned char Base)
/* Read an integer, and store it into IntVal. */
{
unsigned char Val, CharCount = 0;
/* Read the integer value */
IntVal = 0L;
while ((bool) isxdigit (C) && ++Width != 0
&& (Val = HexVal ((char) C)) < Base) {
++CharCount;
IntVal = IntVal * (long) Base + (long) Val;
ReadChar ();
}
/* If we didn't convert anything, it's a failure. */
if (CharCount == 0) {
Error (RC_NOCONV);
}
/* Another conversion */
Converted = true;
}
static void AssignInt (void)
/* Assign the integer value in Val to the next argument. The function makes
** several non-portable assumptions, to reduce code size:
** - signed and unsigned types have the same representation.
** - short and int have the same representation.
** - all pointer types have the same representation.
*/
{
if (NoAssign == false) {
/* Get the next argument pointer */
(void*) __AX__ = va_arg (ap, void*);
/* Put the argument pointer into the zero-page. */
asm ("sta ptr1");
asm ("stx ptr1+1");
/* Get the number of bytes-1 to copy */
asm ("ldy %v", IntBytes);
/* Assign the integer value */
Loop: asm ("lda %v,y", IntVal);
asm ("sta (ptr1),y");
asm ("dey");
asm ("bpl %g", Loop);
/* Another assignment */
asm ("inc %v", Assignments);
asm ("bne %g", Done);
asm ("inc %v+1", Assignments);
Done: ;
}
}
static void __fastcall__ ScanInt (unsigned char Base)
/* Scan an integer including white space, sign and optional base spec,
** and store it into IntVal.
*/
{
/* Skip whitespace */
SkipWhite ();
/* Read an optional sign */
ReadSign ();
/* If Base is unknown (zero), figure it out */
if (Base == 0) {
if (CHAR (C) == '0') {
ReadChar ();
switch (CHAR (C)) {
case 'x':
case 'X':
Base = 16;
Converted = true;
ReadChar ();
break;
default:
Base = 8;
/* Restart at the beginning of the number because it might
** be only a single zero digit (which already was read).
*/
PushBack ();
C = '0';
}
} else {
Base = 10;
}
}
/* Read the integer value */
ReadInt (Base);
/* Apply the sign */
if (Positive == false) {
IntVal = -IntVal;
}
/* Assign the value to the next argument unless suppressed */
AssignInt ();
}
static char GetFormat (void)
/* Pick up the next character from the format string. */
{
/* return (F = *format++); */
(const char*) __AX__ = format;
asm ("sta regsave");
asm ("stx regsave+1");
++format;
asm ("ldy #0");
asm ("lda (regsave),y");
asm ("ldx #>$0000");
return (F = (char) __AX__);
}
int __fastcall__ _scanf (const struct scanfdata* D,
const char* format_, va_list ap_)
/* This is the routine used to do the actual work. It is called from several
** types of wrappers to implement the actual ISO xxscanf functions.
*/
{
register char* S;
bool HaveWidth; /* True if a width was given */
bool Match; /* True if a character-set has any matches */
char Start; /* Walks over a range */
/* Place copies of the arguments into global variables. This is not very
** nice, but on a 6502 platform it gives better code, since the values
** do not have to be passed as parameters.
*/
D_ = D;
format = format_;
ap = ap_;
/* Initialize variables */
Converted = false;
Assignments = 0;
CharCount = 0;
/* Set up the jump "label". CheckEnd() will use that label when EOF
** is reached. ReadInt() will use it when number-conversion fails.
*/
if ((unsigned char) setjmp (JumpBuf) == RC_OK) {
Again:
/* Get the next input character */
ReadChar ();
/* Walk over the format string */
while (GetFormat ()) {
/* Check for a conversion */
if (F != '%') {
/* Check for a match */
if ((bool) isspace ((int) F)) {
/* Special white space handling: Any whitespace in the
** format string matches any amount of whitespace including
** none(!). So this match will never fail.
*/
SkipWhite ();
continue;
}
Percent:
/* ### Note: The opposite test (C == F)
** would be optimized into buggy code!
*/
if (C != (int) F) {
/* A mismatch -- we will stop scanning the input,
** and return the number of assigned conversions.
*/
goto NoConv;
}
/* A match -- get the next input character, and continue. */
goto Again;
} else {
/* A conversion. Skip the percent sign. */
/* 0. Check for %% */
if (GetFormat () == '%') {
goto Percent;
}
/* 1. Assignment suppression */
NoAssign = (F == '*');
if (NoAssign) {
GetFormat ();
}
/* 2. Maximum field width */
Width = UINT_MAX;
HaveWidth = (bool) isdigit (F);
if (HaveWidth) {
Width = 0;
do {
/* ### Non portable ### */
Width = Width * 10 + (F & 0x0F);
} while ((bool) isdigit (GetFormat ()));
}
if (Width == 0) {
/* Invalid specification */
/* Note: This method of leaving the function might seem
** to be crude, but it optimizes very well because
** the four exits can share this code.
*/
_seterrno (EINVAL);
Assignments = EOF;
PushBack ();
return Assignments;
}
/* Increment-and-test makes better code than test-and-decrement
** does. So, change the width into a form that can be used in
** that way.
*/
Width = ~Width;
/* 3. Length modifier */
IntBytes = sizeof(int) - 1;
switch (F) {
case 'h':
if (*format == 'h') {
IntBytes = sizeof(char) - 1;
++format;
}
GetFormat ();
break;
case 'l':
if (*format == 'l') {
/* Treat long long as long */
++format;
}
/* FALLTHROUGH */
case 'j': /* intmax_t */
IntBytes = sizeof(long) - 1;
/* FALLTHROUGH */
case 'z': /* size_t */
case 't': /* ptrdiff_t */
/* Same size as int */
case 'L': /* long double - ignore this one */
GetFormat ();
}
/* 4. Conversion specifier */
switch (F) {
/* 'd' and 'u' conversions are actually the same, since the
** standard says that even the 'u' modifier allows an
** optionally signed integer.
*/
case 'd': /* Optionally signed decimal integer */
case 'u':
ScanInt (10);
break;
case 'i':
/* Optionally signed integer with a base */
ScanInt (0);
break;
case 'o':
/* Optionally signed octal integer */
ScanInt (8);
break;
case 'x':
case 'X':
/* Optionally signed hexadecimal integer */
ScanInt (16);
break;
case 's':
/* Whitespace-terminated string */
SkipWhite ();
CheckEnd (); /* Is it an input failure? */
Converted = true; /* No, conversion will succeed */
if (NoAssign == false) {
S = va_arg (ap, char*);
}
while (C != EOF
&& (bool) isspace (C) == false
&& ++Width) {
if (NoAssign == false) {
*S++ = C;
}
ReadChar ();
}
/* Terminate the string just read */
if (NoAssign == false) {
*S = '\0';
++Assignments;
}
break;
case 'c':
/* Fixed-length string, NOT zero-terminated */
if (HaveWidth == false) {
/* No width given, default is 1 */
Width = ~1u;
}
CheckEnd (); /* Is it an input failure? */
Converted = true; /* No, at least 1 char. available */
if (NoAssign == false) {
S = va_arg (ap, char*);
/* ## This loop is convenient for us, but it isn't
** standard C. The standard implies that a failure
** shouldn't put anything into the array argument.
*/
while (++Width) {
CheckEnd (); /* Is it a matching failure? */
*S++ = C;
ReadChar ();
}
++Assignments;
} else {
/* Just skip as many chars as given */
while (++Width) {
CheckEnd (); /* Is it a matching failure? */
ReadChar ();
}
}
break;
case '[':
/* String using characters from a set */
/* Clear the set */
memset (CharSet, 0, sizeof (CharSet));
/* Skip the left-bracket, and test for inversion. */
Invert = (GetFormat () == '^');
if (Invert) {
GetFormat ();
}
if (F == ']') {
/* Empty sets aren't allowed; so, a right-bracket
** at the beginning must be a member of the set.
*/
AddCharToSet (F);
GetFormat ();
}
/* Read the characters that are part of the set */
while (F != '\0' && F != ']') {
if (*format == '-') { /* Look ahead at next char. */
/* A range. Get start and end, skip the '-' */
Start = F;
++format;
switch (GetFormat ()) {
case '\0':
case ']':
/* '-' as last char means: include '-' */
AddCharToSet (Start);
AddCharToSet ('-');
break;
default:
/* Include all characters
** that are in the range.
*/
while (1) {
AddCharToSet (Start);
if (Start == F) {
break;
}
++Start;
}
/* Get next char after range */
GetFormat ();
}
} else {
/* Just a character */
AddCharToSet (F);
/* Get next char */
GetFormat ();
}
}
/* Don't go beyond the end of the format string. */
/* (Maybe, this should mean an invalid specification.) */
if (F == '\0') {
--format;
}
/* Invert the set if requested */
if (Invert) {
InvertCharSet ();
}
/* We have the set in CharSet. Read characters and
** store them into a string while they are part of
** the set.
*/
Match = false;
if (NoAssign == false) {
S = va_arg (ap, char*);
}
while (IsCharInSet () && ++Width) {
if (NoAssign == false) {
*S++ = C;
}
Match = Converted = true;
ReadChar ();
}
/* At least one character must match the set. */
if (Match == false) {
goto NoConv;
}
if (NoAssign == false) {
*S = '\0';
++Assignments;
}
break;
case 'p':
/* Pointer, general format is 0xABCD.
** %hhp --> zero-page pointer
** %hp --> near pointer
** %lp --> far pointer
*/
SkipWhite ();
if (CHAR (C) != '0') {
goto NoConv;
}
Converted = true;
ReadChar ();
switch (CHAR (C)) {
case 'x':
case 'X':
break;
default:
goto NoConv;
}
ReadChar ();
ReadInt (16);
AssignInt ();
break;
case 'n':
/* Store the number of characters consumed so far
** (the read-ahead character hasn't been consumed).
*/
IntVal = (long) (CharCount - (C == EOF ? 0u : 1u));
AssignInt ();
/* Don't count it. */
if (NoAssign == false) {
--Assignments;
}
break;
case 'S':
case 'C':
/* Wide characters */
case 'a':
case 'A':
case 'e':
case 'E':
case 'f':
case 'F':
case 'g':
case 'G':
/* Optionally signed float */
/* Those 2 groups aren't implemented. */
_seterrno (ENOSYS);
Assignments = EOF;
PushBack ();
return Assignments;
default:
/* Invalid specification */
_seterrno (EINVAL);
Assignments = EOF;
PushBack ();
return Assignments;
}
}
}
} else {
NoConv:
/* Coming here means a failure. If that happens at EOF, with no
** conversion attempts, then it is considered an error; otherwise,
** the number of assignments is returned (the default behaviour).
*/
if (C == EOF && Converted == false) {
Assignments = EOF; /* Special case: error */
}
}
/* Put the read-ahead character back into the input stream. */
PushBack ();
/* Return the number of conversion-and-assignments. */
return Assignments;
}